Pattern Discovery in the Financial Time Series Based on Local Trend

We introduce a method for discovering new patterns in financial time series. Our method focuses on two main tasks of time series mining: Start with time series representation which helps to reduce the dimension and extracts useful feature of raw time series; Next discover on symbolic time series to find out new useful patterns which helpfully to improve trading decision in financial domain. In our work, we are interested in some patterns which have high win ratio percent (i.e. greater 70 %). In the first phrase, (i) raw data will be split into some segments with same length, (ii) local trend will be used to convert each subsequence into symbolic (U, u, s, d or D). In second phrase, we use a sliding window with size w moved on symbolic time series to create a collection of transactions. Based on this collection, the SPAM algorithm is used to discover all patterns with low minSup. In the last phrase, win/loss constraint used to discover new patterns in financial time series will be presented. Our demonstrate based on Gold Spot dataset from 2012-01-01 to 2015-01-01 is experimented.